In theory, discovering the causes of schizophrenia should be simple. First, look for gradients in the incidence of schizophrenia across time or within or between populations. Next, scrutinize the findings carefully and generate candidate risk factors that may explain any detected gradient. Take your top candidate exposures, and undertake an integrated research program involving analytical and genetic epidemiology and applied neuroscience in order to unravel the mechanism of action linking the candidate to schizophrenia. Finally, for modifiable risk factors, undertake randomized controlled trials of interventions designed to ameliorate the exposure in order to confirm the causal link and provide effective preventive interventions. What could be simpler?

In reality, the search for modifiable risk factors for schizophrenia has been a frustrating and often haphazard affair. For many decades, the schizophrenia research community had a scientific scotoma with respect to variations in the incidence of schizophrenia (1). After convincing gradients were identified, the response of the research community was somewhat ataxic. For example, even after a substantial evidence base had accumulated, the default interpretation of the increased risk of schizophrenia in dark-skinned migrants to the United Kingdom was denial—the findings could not be true and must be the result of methodological biases—rather than an attempt to understand a range of possible reasons for the association, which has been done for physical ill-health, such as hypertension, diabetes, and cardiovascular disease. One of the most consistent and long-standing risk factors associated with schizophrenia has been winter-spring birth. However, the general response of the research community has been replication of the ecological association without the subsequent testing of plausible modifiable candidate exposures. Have we wasted time and squandered research opportunities by mishandling these important clues? To be fair, these issues are not unique to schizophrenia research or to psychiatry in general. Apart from the need to follow up clues in a timely and efficient manner, the wider role of observational epidemiology in modern health research has come under scrutiny in recent years; some have suggested that epidemiology has “had its day” (2).

In this issue of the Journal, four articles demonstrate how epidemiology has refined our search space for candidate exposures (3–6). The studies are exemplars of modern epidemiological designs.

Clarke et al. (4) have refined our understanding of the antecedents of schizophrenia. Linking various registers, they have confirmed that infants and toddlers who are in the lower end of the normal distribution for motor milestones have an increased risk of schizophrenia. The results suggest that the processes that underlie the emergence of schizophrenia during young adulthood also affect early motor outcomes. Consistent with a large body of convergent research, these findings suggest that early-life neurodevelopmental processes contribute to the risk of schizophrenia (7). Importantly, in terms of looking for causal risk factors, the authors note that childhood motor delay is “neither a necessary nor a sufficient cause of schizophrenia, and these deficits have weak positive predictive power.” They highlight the importance of their findings in terms of the additive interaction between motor delay and obstetric complications (defined by a low Apgar score). Among 142 case patients and 133 comparison subjects, they found that individuals with both developmental delay and a low Apgar score had greater odds of schizophrenia than would be predicted from adding the two risk factors together (i.e., assuming independence on an additive scale). While some have claimed that such additive interactions are causally informative (8), caution is required in making these claims, particularly because epidemiological history shows that such subgroup analyses rarely replicate (9). Even if further replication were to confirm this interaction, its contribution to causing schizophrenia is likely to be small given that only 15 individuals (5%), including both case patients and comparison subjects (and presumably a smaller proportion of comparison subjects or the general population), had both delayed motor development and a low Apgar score.

In another display of the utility of record linkage studies, Benros et al. (6) report that autoimmune disorders and the presence of any infection are associated with an increased risk of schizophrenia. It is gratifying to see this field of research now solidly underpinned by clues from genetics (e.g., the major histocompatibility complex is associated with risk of schizophrenia in genome-wide association studies [GWAS] [10]). In addition, based on clues from schizophrenia epidemiology, the neuroscience community has discovered previously unrecognized pathways linking altered brain function with maternal and postnatal immune activation (11). Again, Benros et al. are interested in additive statistical interactions; they suggest that the additive interaction between autoimmune disease and infection is supportive of their biologically based hypothesis that exposure to infections (and inflammation) increases the permeability of the blood-brain barrier to autoantibodies. As they note, this interaction has to be treated with caution unless replicated.

It is widely accepted that exposure to traumatic events contributes to many adverse health outcomes, not least of which are psychiatric disorders (12). Within this spectrum of pathology, it is now clear that trauma exposure is associated with psychotic-related outcomes. Heins et al. (5) used a sophisticated design involving case patients, unaffected siblings of case patients, and healthy comparison subjects. The authors were able to reduce the influence of unmeasured residual confounding in order to examine more precisely the association between trauma exposure and psychosis-related outcomes. Now we need to explore why a proportion of individuals exposed to traumatic stress develops psychotic-related outcomes. When is the critical window, and what type of trauma is most “toxic” with respect to psychosis, and how do these exposures interact with other susceptibility factors?

Migrant studies have an important place in epidemiology. These “natural experiments” can help partition the role of “place” in the risk architecture of a disorder while holding genetic factors constant. In particular, they can provide important clues for the identification of modifiable risk factors (e.g., diet, lifestyle, toxins). There is a robust and convincing body of evidence indicating that some migrant groups in some countries have an increased risk of schizophrenia (compared either with native-born individuals or with their compatriots in their country of origin) (13). Stress-related mechanisms have also been implicated in these finding, as it has been proposed that chronic “social defeat” may underpin this association (14).

Veling et al. (3) have been able to address this issue in a thought-provoking study. They examined age at migration and risk of psychosis. This is an area where terminology can be confusing. First-generation migrants are born in one country and move to another country; these individuals can arrive as newborn babies, children, teenagers, or adults. Second-generation migrants are those whose parents were born abroad and who themselves were born in the new country. The act of migration is a stressful process, and for some, the stress is amplified by belonging to an ostracized and easily identified minority group. By examining age at migration in first-generation migrants only, the authors were able to explore whether there was a critical window for exposure to the yet-to-be-identified risk moderating variable. If psychosocial mechanisms related to racism and marginalization only operated during adulthood, one would predict that migrants who arrive as babies and those who arrive as teenagers would be exposed to the same “dose” of psychosocial stress (i.e., throughout adulthood, assuming they do not return, once migrated, to their country of origin). Indeed, those who arrive as young adults might be prone to the greatest social dislocation and stress due to lack of language skills. However, Veling et al. found that migrants who arrive as babies or toddlers had the highest risk of schizophrenia, with the risk decreasing with age at migration thereafter, such that those who migrated at age 29 years or older had no greater risk of psychotic disorder than the nonmigrant population. They conclude that the critical window of exposure is during early life.

This study will trigger many provocative research questions. Do babies and toddlers experience social defeat? Is the mechanism indirect (e.g., stressed parents may change their parenting style and then alter child rearing)? Would we expect that parent-mediated social defeat operating in babies and toddlers would result in similar adult mental health outcomes compared with social defeat experienced by children and young adults? Might the risk associated with second-generation migrants be “imprinted” during early life also? Is schizophrenia a long-latency outcome for certain types of early-life stress exposures?

Clues from epidemiology have revitalized social psychiatry research in schizophrenia; now we need to explore a taxonomic approach to determine what types of stress affect individuals during which phases of the lifespan. Which exposures are best assessed at the individual level and which at the ecological (family, neighborhood, city, etc.) level? The epidemiology research community has learned important lessons in recent years about the dangers of strong inferences from observational epidemiology (15). We are in danger of misidentifying proxy risk indicators (not in the causal pathway) with modifiable causal risk factors. In particular, exposures may relate in a complex, socially patterned fashion at the individual and family levels such that the putative risk factors may not be in the causal pathway. Trauma exposures rarely occur in isolation (12). Could migrant status be a proxy marker for an unmeasured exposure (e.g., low vitamin D, infection)?

Epidemiology can offer other ways to explore the association between candidate exposures and schizophrenia. Germ-line genetic variation—for example, single nucleotide polymorphisms (SNPs)—can be used as proxy (“instrumental”) markers to obtain causal estimates of the effects of modifiable risk factors on disease outcomes (e.g., SNPs that affect enzymes involved in folate metabolism may serve as proxies for serum folate concentrations). This method, termed Mendelian randomization, has been shown to be less biased and confounded than conventional multivariable regression approaches and provides an interesting addition to the epidemiology tool kit (16). Sometimes these clues can emerge from genetic studies; for example, a large GWAS for multiple sclerosis identified SNPs in two different enzymes involved in vitamin D metabolism (17), providing some support for the hypothesis that low vitamin D is a causal risk factor for this condition. Observational epidemiology (including migrant studies) and animal research have long implicated these pathways in this disorder.

With respect to prevention, are any of the modifiable risk factors linked to schizophrenia ready for randomized controlled trials? Can we provide intervention for at-risk groups (e.g., children exposed to trauma, dark-skinned migrants in certain countries) (18)? While candidate exposures related to infection and nutrition may appear to be the most pragmatic for public health interventions (e.g., vaccines, nutritional supplementation), the evidence base is not sufficient to justify these interventions yet. With respect to area-level psychosocial stress factors and childhood trauma, are there pragmatic interventions that can reduce these exposures? Should we wait for the evidence to accumulate, or if the proposed intervention is inexpensive and safe, should we move toward randomized controlled trials promptly (19)? Based on high-quality multisite incidence studies, Kirkbride et al. (20) have estimated that up to 22% of all psychoses in England could be prevented if the yet-to be-identified migrant-related exposures could be removed. Within the black minority groups, the proportion is two-thirds of all cases. These estimates cannot be ignored and are impressive in comparison to the prevalence of some of the other (additively combined) risk factors examined in the series of articles in this issue.

In the last two decades, substantial progress has been made in understanding the epidemiology of schizophrenia. We now have candidate exposures in our sights or at least a more manageable search space from which to generate our candidates. The new findings should be used to sharpen our hypotheses and design the next generation of studies to refine our understanding of the modifiable risk factors for schizophrenia.